PERVERSE COHERENT SHEAVES DMITRY ARINKIN AND ROMAN BEZRUKAVNIKOV Abstract. This note introduces an analogue of perverse t-structure [1] on the derived category of coherent sheaves on an algebraic stack (subject to some mild technical conditions). Under additional assumptions construction of coherent “intersection cohomology” sheaves is given. Those latter assumptions are rather restrictive but hold in some examples of interest in representation theory. Similar results were obtained by Deligne (unpublished), Gabber [10] and Kashiwara [13]. To Pierre Deligne with admiration. 1. Introduction Let X be a reasonable stratified topological space; or let X be a reasonable scheme, stratified by locally closed subschemes. Let D be the full subcategory in, respectively, derived category of sheaves on X, or in the derived category of ´etale sheaves on X, consisting of complexes smooth along the stratification. For an integer-valued function p (perversity) on the set of strata Beilinson, Bern- stein, and Deligne [1] defined a t-structure on the category D; the objects of corre- sponding abelian category (core of the t-structure) are called perverse sheaves. The question addressed in this note is whether an analogous construction can be carried out for the derived category of coherent sheaves on a reasonable scheme. Surprisingly, the answer is positive (with some modifications), easy, and not widely known (although it was known to Deligne for a long time, see [8]). Let us summarize the difference between the coherent case considered here, and the constructible case treated in [1]. First, in the coherent case we can not work with complexes “smooth” along a given stratification, because the corresponding subcategory in the derived category arXiv:0902.0349v3 [math.AG] 3 Feb 2021 is not a full triangulated subcategory. (If f is a function whose divisor intersects f the open stratum, then the cone of the morphism O → O has singularity on the open stratum). This forces us to define perversity as a function on the set of generic points of all irreducible subschemes, i.e. on the topological space of a scheme. The second, more essential difference is that in the derived category of coherent sheaves the functor j∗ of pull-back under an open embedding j does not have adjoint functors. Recall that in constructible situation the right adjoint to j∗ is the functor j∗ of direct image, and the left adjoint is the functor j! of extension by zero. In coherent set-up the functor j∗ is defined in the larger category of quasi-coherent sheaves (Ind-coherent sheaves), while j! is defined in the Grothendieck dual category (consisting of Pro-coherent sheaves) introduced in Deligne’s appendix to [12]. 2020 Mathematics Subject Classification. Primary 18F20; Secondary 14A20, 14F05. Key words and phrases. Coherent perverse sheaves, coherent IC sheaves. 1 2 DMITRYARINKINANDROMANBEZRUKAVNIKOV It turns out, however, that in the proof of the existence of perverse t-structure (one can use instead of the object j!(F) (where j : U ֒→ X is an open embedding any extension F˜ of F to X such that the restriction of F˜ to X −U has no cohomol- ogy above certain degree (depending on the perversity function). If the perversity function is monotone (see Definition 3.9 below) it is very easy to construct such F˜. Applying the Grothendieck-Serre duality to this construction, we get a substitute for j∗(F), which exists if the perversity function is comonotone. Otherwise the proof is parallel to that in [1]. Thus the t-structure is constructed not for an arbitrary perversity function, but only for a monotone and comonotone one. (In the topological situation one also needs this condition to get a t-structure on the whole derived category of con- structible sheaves rather than on the category corresponding to a fixed stratifica- tion.) In [8] Deligne used the Grothendieck’s Finiteness Theorem ([11], VIII.2.1) to p p show that the formulas for τ≤0, τ≥0 of [1], a priori making sense in a larger cate- gory containing Db(Coh), give in fact objects of Db(Coh), provided the perversity function is monotone and comonotone (see also Remark 3.13). The results on the existence of a “perverse” t-structure carry over to the case of G-equivariant coherent sheaves, where G is a (reasonable) algebraic group acting on a (reasonable) scheme X. In this case perversity p(x) is assigned only to orbits of G (including “generic orbits”), since an equivariant sheaf is automatically smooth along the orbits. More generally, the perverse t-structure can be constructed for coherent sheaves on a (reasonable) algebraic stack X , given a perversity defined on points of X . The case of G-equivariant coherent sheaves on X corresponds to working with the stack X = X/G. The general formalism in all three situations: sheaves on a scheme, equivariant sheaves on a scheme, and sheaves on a stack — is very similar, to the extent that we found it easier to work with algebraic stacks and treat the other situations as special cases. However, one construction does not apply to (non-equivariant) sheaves on a scheme. Namely, the definition of the minimal (Goresky-MacPherson, or IC) extension functor j!∗ requires a strictly monotone and comonotone perversity. Such perversity exists only if the dimensions of adjacent points differ by at least two, which excludes schemes (other than finite schemes). On the other hand, in equivariant settings, it is possible that the dimensions of adjacent orbits differ by at least two, and a strictly monotone and comonotone perversity exists. If the perversity is strictly monotone and comonotone, an analogue of the usual description of irreducible perverse sheaves as minimal extensions of local systems is valid, and the core of the perverse t-structure is Artinian and Noetherian (in contrast with the core of the standard t-structure). Some examples of this situation are given at the end of the paper. A version of the main result in a restricted generality appeared in the preprint [3] by the second author. Later related constructions were published by Gabber [10] and Kashiwara [13]. A similar result was known to Deligne [8] long before the date of [3]. Organization. In Section 2, we prove some basic properties of coherent sheaves, and study dualizing complexes on stacks. 3 Section 3 contains the definition of the perverse t-structure on a stack. The main result of this section (Theorem 3.10) verifies axioms of a t-structure. In Section 4, we define the minimal extension functor (Theorem 4.3). We then use it to study irreducible perverse coherent sheaves (Proposition 4.11), and prove that the category of perverse coherent sheaves is Artinian and Noetherian (Corol- lary 4.13). As we already mentioned, these results require additional assumptions; in particular, all results are empty in the case of (non-equivariant) coherent sheaves on a scheme. Acknowledgements. We are much obliged to Pierre Deligne for valuable expla- nations, comments on the text and a kind permission to use his unpublished results. We thank Alexander Beilinson, Victor Ginzburg, and Dmitry Panyushev for discussions and references. This paper grew out of the preprint [3] which was started in the spring of 1999 when the second author was a member at the Institute for Advanced Study. He thanks IAS for excellent work conditions and Leonid Positselski for his participation in the early stages of the work on [3]. The first author is a Sloan Research Fellow, and he is grateful to the Alfred P. Sloan Foundation for the support. The work of the second author was partly supported by DARPA grant HR0011-04-1-0031 and NSF grant DMS-0625234. 2. Preliminaries In this section we collect some results needed in the exposition. 2.1. Quasi-coherent sheaves on stacks. Let X be an algebraic stack (an Artin stack). Suppose X is Noetherian, and in particular, quasi-compact. Assume also that X is semi-separated: that is, the diagonal morphism X →X×X is affine. Let us fix a presentation of X , that is, a surjective smooth morphism π : X → X , where X is an algebraic space. Lemma 2.1. Let X be an algebraic stack. The following conditions are equivalent: (a) X is quasi-compact and semi-separated. (b) X admits a presentation π : X → X such that X is an affine scheme and π is an affine morphism. (c) X admits a presentation π : X → X such that X is a quasi-compact semi- separated algebraic space and π is an affine morphism. Proof. (a)⇒(b). Let π : X → X be a presentation of X . Note that X is quasi- compact, so passing to its ´etale cover, we may assume that X is an affine scheme. But then X ×X X = (X × X) ×(X×X ) X is an affine scheme; therefore, π is an affine morphism. (b)⇒(c) is obvious. (c)⇒(a). Clearly, X is quasi-compact; let us prove it is semi-separated. The morphism X ×X X → X is affine, because it is obtained from π by a base change. Since X is semi-separated, so is X ×X X. Therefore, the composition X ×X X → (X ×X X) × (X ×X X) → X × X 4 DMITRYARINKINANDROMANBEZRUKAVNIKOV is affine. It remains to notice that the composition is obtained from the diagonal X→X×X by a base change. Remark 2.2. To simplify the exposition, we only consider presentations π : X → X where X is a scheme from now on. This assumption allows us to avoid a separate discussion of perverse coherent sheaves on algebraic spaces. Also, we do not need smoothness of presentations: it is enough to assume that π : X → X is a faithfully flat Gorenstein morphism of finite type.